> ## Documentation Index
> Fetch the complete documentation index at: https://docs.sentivue.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Groq

> Connect Groq to SentiVue Talk

Groq runs open models and Whisper on its own inference hardware, which makes it one of the faster options for both. In SentiVue Talk it drives the **language model** and **transcription**.

## What you need

| Field   | Required |
| ------- | -------- |
| API key | Yes      |

One key covers both services.

## Getting your API key

1. Sign in at [console.groq.com/keys](https://console.groq.com/keys).
2. Create a new key and give it a name.
3. Copy the key and store it somewhere safe before leaving the page.

<Warning>
  Groq rate-limits per model, per organization, and the ceilings on the free plan are low. Limits are counted separately for requests per minute, tokens per minute and — for transcription — audio seconds per hour. Hitting any one of them fails the request even when the others have headroom. Check your plan's limits for the specific models you intend to use before putting a number on how many calls you can run at once.
</Warning>

## Adding the connection

1. Go to **Connections** and choose **Groq**.
2. Paste the API key.
3. Choose which services this connection drives — language model, transcription, or both — and save.

SentiVue Talk validates the key with Groq as you save, so a bad or revoked key is caught immediately rather than on a live call.

## What you can select afterwards

Choose your models on [Model Configurations](/talk/configurations/inference-providers).

**Language model** — short identifiers, no vendor prefix. Browse them on Groq's [models page](https://console.groq.com/docs/models), which also marks production versus preview. Keep production agents on a production model — preview ones can be withdrawn at short notice.

**Transcription** — set the language to match your callers rather than leaving Whisper to guess. The [models page](https://console.groq.com/docs/models) lists the current speech models.

## Using Groq for only one service

The connection does not have to drive both roles. Turn off the one you do not want, and SentiVue Talk uses your other connections for it — Groq for transcription with the language model on [OpenAI](/talk/configurations/providers/openai), or Groq for the language model with transcription on [Deepgram](/talk/configurations/providers/deepgram). Change this at any time on the connection.

## Completing the pipeline

Groq has no voice. You also need a [voice provider](/talk/configurations/voice) to speak the agent's replies, as a separate [connection](/talk/configurations/connections).

## Troubleshooting

<AccordionGroup>
  <Accordion title="The key is rejected when I save">
    Confirm you copied the whole key, with no trailing space, and that it has not been deleted or rotated in the Groq console. If you are unsure which value is current, create a new key.
  </Accordion>

  <Accordion title="The agent goes silent partway through a call">
    This is the usual sign of a rate limit. Groq returns `429 Too Many Requests` once you cross a limit, and mid-call there is nothing to fall back on, so the agent simply stops responding until the window resets. Limits are per model, so a busy language model can throttle while transcription is fine, or the reverse. Concurrent calls share the same organization limits — several calls at once will hit them far sooner than testing alone suggests. Raise your plan's limits, or move one of the two roles to another provider.
  </Accordion>

  <Accordion title="A model I entered is not found">
    Check the identifier against Groq's [models page](https://console.groq.com/docs/models). Groq retires models on a fast cycle, and preview models in particular are removed with little notice — a model name that worked last month may no longer exist. Deprecated entries are listed on the same page.
  </Accordion>

  <Accordion title="Transcripts are poor or empty">
    Check the language setting first — a mismatch between the configured language and what the caller speaks is the usual cause. If the audio is noisy or heavily accented, move from the turbo model to full Whisper large-v3, which is more accurate at the cost of some speed.
  </Accordion>
</AccordionGroup>